Hi,
I agree, this should be a change in 9.0 only. In general we can start with web.xml still existing (the bootstrap code would use a Jetty “WebAppContext” initially). At a later stage we can replace the WebappContext by a ServletContextHandler (which is the superclass and offers no webapp anymore) and we can configure everything on our own. Another idea would be to allow the end user to suply his own web.xml somewhere in a directory, where he can add extra filters if really needed. On bootstrap we setup our own WebappContext and tell it to load the user’s web.xml, but linking our own servlets and filters and assets would be done in our code. I don’t really like this approach, because this would disallow to replace jetty by async Netty in the future. Uwe ----- Uwe Schindler Achterdiek 19, D-28357 Bremen https://www.thetaphi.de eMail: u...@thetaphi.de From: Mike Drob <md...@apache.org> Sent: Wednesday, March 25, 2020 6:07 PM To: dev@lucene.apache.org Subject: Re: [DISCUSS] SIP-6 Solr should own the bootstrap process I think we like to pretend that Jetty and web.xml are implementation details, but I'm not sure how good of a job we do. Whether we need to provide backwards compatibility or not is the wrong question I think, because when we break compatibility we are telling a subset of our users that it will not be easy for them to upgrade. I don't know who out there has customizations in web.xml (maybe security filters is the most common?). I don't know what the migration path for those users would look like if they want to add stuff to an embedded jetty that we're running - do we need to provide system properties they can set? Do we need to give them a separate xml file that they can use similar syntax for? I'm definitely against changing that for 8.6, and as much as I want this change for 9.0 I really want to see an alternative that makes sense proposed before we start tearing through all of this. I'll spend a bit more time thinking about this and see if I can come up with any ideas as well. Mike On Wed, Mar 25, 2020 at 11:29 AM Jan Høydahl <jan....@cominvent.com <mailto:jan....@cominvent.com> > wrote: I agree. I changed the SIP page and now have four bullet points defined as in scope for the SIP: The most important change in this SIP will be * Create a solr/bootstrap module with a solr.jar to replace start.jar as the entrypoint (java -jar solr.jar ...) * Hide jetty even more, i.e. rename jetty.port → port , and perhaps clean up other Sysprops as well * Get rid of web.xml and instead build our contexts and paths from Java code, driven by Solr config * Be able to lock Solr's memory to prevent swapping (SOLR-14335) Not sure how big of a job the web.xml thing is, but there is definitely a potential of unifying test and production code in this area and thus get much better test coverage of some of this. However, removing web.xml is perhaps such a big change that we’ll have to target 9.0 while my initial goal of just creating a bootstrapper to lock memory could have been merged to 8.x as well?? Or do we consider web.xml an implementation detail for which we don’t need to provide backward compatibility? I’m a bit back and forth with myself on this. Perhaps if we reduce scope to just the bootstrap module, mem locking and some sysprop renames, it can be done quickly and end up in 8.6 already, then increment from there? Jan 25. mar. 2020 kl. 16:13 skrev Mike Drob <md...@apache.org <mailto:md...@apache.org> >: I'm in favor of this approach mainly because of the line mentioned in the Test Plan: > If we move servlet initialization into Java code and get rid of web.xml, we > may be able to start Solr in the same way both in tests and in real life! I feel that this should be a top level goal, along with taking another look at EmbeddedSolrServer (perhaps it moves out of SolrJ?). Right now, I feel like we effectively have no testing of the configurations that go into the various xml files - can't speak for others, but it made me reluctant to make changes to them. Mike On Tue, Mar 24, 2020 at 6:36 PM Jan Høydahl <jan....@cominvent.com <mailto:jan....@cominvent.com> > wrote: This SIP has changed name to SIP-6 and I have changed the email subject of this thread, please keep this new subject going forward. New SIP link: https://cwiki.apache.org/confluence/display/SOLR/SIP-6+Solr+should+own+the+bootstrap+process Jan 24. mar. 2020 kl. 22:48 skrev Uwe Schindler <u...@thetaphi.de <mailto:u...@thetaphi.de> >: Of course it's a single jvm. It's only a replacement for start.jar. the mock-up is already there. Uwe Am March 24, 2020 9:45:33 PM UTC schrieb Bram Van Dam <bram.van...@intix.eu <mailto:bram.van...@intix.eu> >: On 24/03/2020 16:01, Jan Høydahl wrote: It has been proposed many times to make Solr a truly standalone app and that we should start Jetty from within Solr and not the other way around :) Do you see this running in a single JVM, i.e. you run solr.jar and it then starts an embedded Jetty server? Or would solr.jar exec a new Java process and take it from there? If it's all in one JVM, then the startup scripts probably won't be all that different. Memory settings, OOM killers etc will still have to be set up before using startup shellscripts and whatnot. If it's just a small bootstrap jar which execs another JVM, a lot of the complicated shell foo could probably be simplified. - Bram _____ To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org <mailto:dev-unsubscr...@lucene.apache.org> For additional commands, e-mail: dev-h...@lucene.apache.org <mailto:dev-h...@lucene.apache.org> -- Uwe Schindler Achterdiek 19, 28357 Bremen https://www.thetaphi.de <https://www.thetaphi.de/>
